EM

If we want to find the maximal point of x for the function f(x), we can introduce another parameter y to help

xmaxargmaxxf(x)

take f(x) apart into

f(x)g(x,y)+Δ(x,y)ykargmaxyg(xk,y)argminyΔ(xk,y)

where yk is easy to find, minyΔ(xk,y)=0

step 1 [M]

fix yyk

xk+1argmaxxg(x,yk)g(xk+1,yk)>g(xk,yk)=f(xk)

step 2 [E]

fix x=xk+1

yk+1argmaxyg(xk+1,y)argminyΔ(xk+1,y)=0f(xk+1)g(xk+1,y)+Δ(xk+1,y)=g(xk+1,yk)+Δ(xk+1,yk)=g(xk+1,yk+1)+Δ(xk+1,yk+1)=g(xk+1,yk+1)f(xk+1)=g(xk+1,yk+1)>g(xk+1,yk)

In all

f(xk+1)=g(xk+1,yk+1)>g(xk+1,yk)>f(xk)=g(xk,yk)

So

limkf(xk)=limkg(xk+1,yk)=argmaxxf(x)

Comment:

By introducing another parameter y, and iterate step 1 [M] and step 2 [E], we eventually find the maximal point of x for the function f(x)